Don Chmielewski and Anna Tong
Scarlett Johansson on Monday accused OpenAI of creating a voice for the ChatGPT system that sounded “eerily similar” to the actress’s after she herself refused to voice the chatbot.
Johansson made the comments in a statement released hours after the artificial intelligence company said it was removing a voice called “Sky.”
OpenAI CEO Sam Altman said in a statement emailed to Reuters on Monday that Skye’s voice was not an imitation of Johansson, but that of another professional actress.
“Skye’s voice is not Scarlett Johansson’s and was never intended to resemble her voice. We selected a voice actor to do Skye’s voice before turning to Ms. Johansson,” Altman said.
“Out of respect for Ms. Johansson, we have suspended the use of Skye’s voice in our products. We are sorry to Ms. Johansson that we could not have communicated better.”
The fight over the rights to the voices and images of actors has become the focus of attention in Hollywood as studios consider how to use artificial intelligence to create new entertainment and computer-generated images and sounds become increasingly difficult to distinguish from human ones.
Johansson said in a statement that Altman approached her last September and offered to hire her to provide the voice of ChatGPT – an offer she declined.
“Nine months later, my friends, family and the general public have commented on how much the newest system called Sky is like me,” she said.
“When I heard the demo released, I was shocked, angry and in disbelief that Mr. Altman would pursue a voice that was so similar to mine that my closest friends and news outlets could not tell the difference.”
Johansson added that Altman “hinted that the similarity was intentional” by tweeting a link to “Her,” a 2013 film about a man who develops a relationship with an artificial intelligence assistant voiced by the actress.
Johansson’s note was published by journalists from NPR and other news agencies. Her publicist also shared this with Reuters.
She said she hired an attorney to learn about the process of creating the voice.
Last week, OpenAI showed off its latest AI model, called GPT-4o, with audio capabilities that allow users to talk to a chatbot and receive responses in real time, marking significant progress in making AI conversations sound more realistic.
(This story has been rewritten to remove an earlier link to OpenAI, which did not immediately provide a comment in paragraph 12)